Michael Woods blew away the room at OpRiskNA last month with his discussion on how Machine Intelligence helps construct effective revenue forecast models for CCAR. While we talk about the subject in some detail below, the video of the talk and the superb Q&A that follows is not to be missed. Check it out here:
Feel free to grab the slide deck here.
After the market collapse in 2008, the Federal Reserve implemented a series of annual tests to ensure that the largest bank holding companies in the US could withstand another potential economic and financial shock. The tests, DFAST and CCAR, assess the bank’s capabilities based on adverse economic scenarios relying on their data, requirements, and supervisory exercises.
The toughest issues in terms of operational risk management include model creation, model validation, and model governance. Simultaneously, the growing issue of data complexity makes this increasingly challenging because as it grows, it also becomes more and more difficult to build effective models that are statistically valid.
With these data sets, even the simplest model construction procedure produces a massive number of potential models that you need to consider. To put things into perspective, a “simple” data set with thousands of variables has over two trillion models. Within this immense amount, you also have to keep in mind that many are invalid and will be no help in effectively forecasting revenue.
In order to weed out and assess these variables, you must combine machine intelligence and human intelligence to drill down to exactly which of those two trillion models are necessary to explore.
Ayasdi has a software platform for executing and developing Machine Intelligence applications that leverage complex data. We have the ability to apply a framework that consolidates this algorithmic information and identifies the most important structure in the data. You are then able to identify which algorithmic information is most important to your dataset and present it to the user in a way that allows you to really capture and convey the complexity.
Michael’s talk outlines how Ayasdi helped a major US bank rapidly identify those variables that impact revenue, from thousands of variables, and then create effective forecast models that helped them pass CCAR with flying colors.
The story here is distilling complexity – something our software excels at. For this client, and others in the financial services space, our ability to find the elements that are important within complex, highly-dimensional data sets is what distinguishes us from the “point solutions” that abound. That’s important because with 40 percent growth in data volume, the number of potential models for the client discussed in this video will exceed 8 trillion in the next year – and that’s a lot.